Rule 34 Hot Logo
Rule 34 Hot Logo

Legal

AI Image Generation

How to use AI to generate custom images — characters, realistic, triggers, prompts, and settings.

In this article

Generate custom AI images in three steps:

1

Find a Model

Search for a character, style, or concept

2

Write a Prompt

Describe what you want to see

3

Generate

Press Generate and get your image

What Is AI Image Generation?

AI image generation creates pictures from text descriptions. You type what you want to see (called a prompt), and the AI draws it for you. The results are unique every time -- even the same prompt produces different images.

You pick a model (a style or character add-on), write your prompt, adjust optional settings, and the AI generates your image in seconds.

What Are AI Models?

An AI model is a small add-on trained on a specific subject -- a character, an art style, a concept, or a pose. Think of it as teaching the AI something new.

For example, a model trained on a specific anime character knows exactly how that character looks -- their hair, eyes, outfit, proportions. Without the model, the AI would just guess based on your text description.

We have two generators:

- Rule 34 Characters -- anime, hentai, game and movie characters (Illustrious and Pony models) - Realistic AI Porn -- photorealistic images with styles, positions, and concepts (Pony and SDXL models)

Both produce high-quality results. Choose the one that matches the type of content you want to create.

Finding the Right Model

Go to the Generate page from the top navigation menu. Type a character name, franchise, art style, or any keyword in the search bar.

Results show as cards with preview images. Click any card to open the model detail page where you can see example images, read the description, and start generating.

The search is smart -- it scores results by how well they match your query, prioritizing exact name matches, popular models, and models with more likes.

Understanding Trigger Words

Trigger words are special keywords that activate a model. When a model was trained, certain words were associated with its content. Including these words in your prompt tells the AI: "use this model's training right now."

For example, a character model might have a trigger word like "hinata_hyuga" or "nami_one_piece". Adding this to your prompt helps the AI generate that specific character accurately.

Trigger words appear as clickable buttons on the model page. Click them to add them to your prompt.

Not All Models Need Triggers

Some models have no trigger words at all and work just fine -- the training is activated automatically. This is common for style models and some character models.

Other models have trigger words but still produce decent results without them. The triggers just make the output more accurate and consistent.

As a rule of thumb:

- Character models usually benefit the most from trigger words. They help the AI nail the character's appearance. - Style models often work without triggers. The style is applied regardless of your prompt. - Concept models (poses, settings, themes) may or may not need triggers depending on training.

If a model has trigger words and you're not sure whether to use them, try generating with them first. You can always remove them and try again.

Writing Good Prompts

Your prompt is the text description of what you want the AI to generate. Better prompts produce better images. Here are some tips:

Be specific. Instead of "a girl on a beach", try "1girl, standing on sandy beach, blue bikini, ocean waves, sunny day, blue sky".

Use tags, not sentences. AI models understand comma-separated tags better than natural language. "1girl, red hair, green eyes, school uniform" works better than "a girl who has red hair and green eyes wearing a school uniform".

Include quality tags. Adding tags like "masterpiece, best quality, highly detailed" at the beginning of your prompt often improves results.

Use negative prompts. The negative prompt tells the AI what to avoid. Common negative tags include "low quality, blurry, bad anatomy, extra limbs, deformed". These help prevent common AI artifacts.

Using Example Images

Most models come with example images that show what the model can do. These are powerful because they include the exact settings, prompt, and trigger words that were used.

Click any example image to automatically load its prompt, negative prompt, and all generation settings (steps, CFG scale, scheduler, dimensions) into the form. This is often the best way to start -- you get proven settings, then tweak the prompt to create what you want.

Example images only show up if the model creator provided them with generation metadata.

Generation Settings

You can use the default settings for most generations. For more control, here is what each setting does:

Steps (10-50): How many refinement passes the AI makes. More steps = more detail, but slower. 20-30 is usually enough.

CFG Scale (1-15): How closely the AI follows your prompt. Higher = stricter adherence but can look unnatural. 5-8 works well for most cases.

Scheduler: The algorithm used for generating. "Euler A" is the default and works great. Different schedulers can produce subtly different styles.

Size presets: Choose between Portrait, Landscape, Square, Wide, and Tall formats. The default is Portrait which works well for character images.

Seed: A number that determines the randomness. Set to -1 (random) by default. Use a specific seed to reproduce the exact same image with the same prompt and settings.

Model Strength (in Advanced): How strongly the model is applied. Default is 0.8. Higher values (up to 1.5) make the model's style or character more dominant, but can also cause artifacts.

Clip Skip (in Advanced): Controls which layer of the text encoder is used. Usually 2 for anime models. Rarely needs changing.

Base Checkpoints

Each model runs on top of a base checkpoint -- the core AI model that does the actual image generation. You can choose between:

- Default: Optimized for anime and illustration style. Good all-around choice. - Realistic: Produces more photorealistic results. Best for models trained on realistic content.

The checkpoint selector appears on the model page if options are available. When in doubt, stick with Default.

Model Variations

Some models have multiple versions (e.g. "Ver 1", "Ver 2"). These may be trained on different base models (Illustrious vs Pony) or represent updated training data. Try different versions to see which style you prefer -- they can produce quite different results.

The Generation Process

After you press Generate:

1. Your request is sent to the server and queued 2. The AI begins creating your image (you'll see a "Generating" spinner) 3. When complete, your image appears in the generation panel 4. You can download the image or click Generate Again for a new one

Generation typically takes 10-60 seconds depending on server load. You can leave the page and come back -- your generation is saved and will resume automatically.

When Generation Fails

Occasionally a generation may fail. Common reasons:

- Server busy: High demand can cause timeouts. Try again in a moment. - Content filter: Some prompts may be rejected by the generation network. - Model unavailable: Rarely, a specific model version may be temporarily offline.

When a generation fails, your energy is automatically refunded. Just adjust your prompt or try again.

Tips for Better Results

1. Start with examples. Click an example image to load proven settings, then change the prompt. 2. Use trigger words for character models -- they make a big difference in accuracy. 3. Add quality tags like "masterpiece, best quality" at the start of your prompt. 4. Use negative prompts to avoid artifacts: "low quality, bad anatomy, blurry, extra limbs". 5. Try different seeds. If you like a composition but want variation, keep the prompt and change the seed. 6. Adjust model strength if the character looks "too much" (lower it) or "not enough" (raise it). 7. Check model descriptions. Many creators include recommended prompts and tips on their model pages.

HD Upscaling

Before publishing, every image must be upscaled to HD quality. The upscaler takes your generated image and enhances it to 4K resolution (approximately 2x the original dimensions -- for example, 832x1216 becomes 1664x2432).

The upscaler does more than just increase the size:

- Face fixing: Corrects distorted or blurry faces, a common AI artifact - Detail enhancement: Sharpens fine details like hair strands, fabric texture, and background elements - Artifact removal: Reduces noise, broken fingers, and other generation flaws - Smart upscaling: Uses the original generation parameters (prompt, seed, model) to guide the upscaling for better results

To upscale, click the "HD" button on any completed image, or use "Upscale All HD" to upscale the entire batch at once. Upscaling takes about 10-30 seconds per image. A green "HD" badge appears on the image when complete.

How the Upscale Queue Works

Images are upscaled one at a time through a global queue. When you submit an image for upscaling, you'll see its position in the queue (e.g. "Position: 3"). The status progresses through:

- Queued (blue): Waiting in line - Upscaling (orange): Currently being processed - HD (green): Complete -- the image is now in 4K - Failed (red): Something went wrong -- click "Retry" to try again

You can continue generating new images or browse the site while waiting. The upscaled image automatically replaces the original in the generation panel once complete.

Publishing to the Platform

Once your image is upscaled to HD, you can publish it as a post on the platform. Here's the full flow:

1. Generate 1-4 images from a model 2. Upscale the image(s) you like to HD quality 3. Choose one image to publish -- only one image per generation batch can be published 4. Click "Publish" on your chosen image 5. Add tags: At least one character tag is required. Copyright/series tags are optional. 6. Set rating: Explicit, Questionable, or Safe 7. Submit: The image is optimized, uploaded, and created as a post

Prompt tags from your generation are automatically matched against our tag database and added to the post. AI meta tags (like "ai_generated") are added automatically so users know the image was AI-created.

One Publish Per Batch

You can only publish one image from each generation batch. This keeps the platform focused on quality over quantity -- pick your best result from each batch.

Once you publish one image, the other images in that batch show "Limit Reached" and cannot be published. You can still download any image. To publish more images, simply generate a new batch and pick the best one again.

Publish Requirements

To publish an AI-generated image, all of these must be met:

1

HD Upscaled

Image must be upscaled to 4K first

2

Character Tag

At least one character tag is required

3

No Prohibited Content

Prompt must not contain prohibited terms

4

One Per Batch

Only one image per generation batch

Energy Costs for Generation

Each AI image generation costs 100 energy. Generating a batch of 2-4 variations costs 100 energy per image. HD upscaling also costs 100 energy per image. Failed generations and upscales are automatically refunded.

New accounts receive 100 free energy (one free generation). After that, purchase more via packages starting at $6 (with 25% sitewide promo). See our Energy & Pricing guide for full details on packages and pricing.
Back to Help Center